Monitoring

Bounded error aggregation

Configure ErrorTracker with retentionMs for a fixed window starting at first occurrence, optional fingerprint/redaction functions, logEntries: false for silent aggregation, and onNewError for host alert policy. Expiry is applied on reads and writes; no recurring timer or lifecycle provider is needed. errorFingerprint hashes the error name/message/first stack location and endpoint. Default consumers retain the original category/message aggregation and unlimited retention. External reporting remains behind ErrorReporter; product classifications, critical keywords and tracking enablement stay in the host.

Optional Sentry adapter and selected health probes

SentryErrorReporter receives explicit SDK initialization options, selected redaction fields/sections and a context mapper. It loads @sentry/node only when configured and initializes once. The application owns its enablement decision, DSN, sampling, tags and sensitive field list. ERROR_REPORTER can still use NoopErrorReporter.

HealthCheckService.checkAll(names?) reuses the existing status aggregation for readiness or diagnostics subsets. A requested but unregistered probe is DOWN. HTTP paths, status field naming and business metric queries stay with the application. MetricsRegistry plus renderPrometheusMetrics provide serialization without a second text renderer.